When companies must report carbon emissions using the same methodology, for example, or disclose labor conditions using identical categories, investors can compare performance, identify laggards and allocate capital accordingly. This push made comparisons possible and deception harder, although it still wasn’t foolproof. For example, a 2023 United Nations Environment Programme report on the fast-fashion industry found that many companies continue to make “vague and inflated” sustainability claims.

Second, there are no comprehensive frameworks in the U.S. that require businesses to judge how AI affects them in a material way and publicly disclose those impacts. Examples of AI-driven material impacts include whether algorithmic bias shapes business outcomes, or whether decisions on how to use AI systems carry significance for shareholders and the public.

Instead, AI governance remains dominated by the narrow inner circle of companies that build the AI systems, while affected communities rarely have a say in determining which AI impacts are material enough to warrant public attention. For example, Big Tech companies like Google, Microsoft, Apple, NVIDIA and others adhere to their own AI governance guidelines, with relatively little public input.

The development of sustainability principles offers some examples of how to build these frameworks. The EU’s Corporate Sustainability Reporting Directive requires over 50,000 companies to formally evaluate which sustainability topics are material to their stakeholders, and then disclose that information. These efforts try to ensure that accountability is clear across entire supply chains.

While nowhere nearly as comprehensive, U.S. regulations such as the 2010 Dodd-Frank financial reform and California’s law requiring reporting on statewide greenhouse gas emissions provide a similar blueprint that U.S. policymakers could build on if they chose.

A third problem is the general lack of third-party verification, making AI washing trivially easy. Effective disclosure means reporting all material impacts – not just cherry-picked successes.

In practice, AI audits can vary dramatically in rigor, scope and methodology. One auditor might conduct extensive testing across demographic groups, analyze decision-making and validate the quality of training data. Another might simply review documentation and accept company explanations at face value. Given the variety of AI auditing models out there, different auditors may use incompatible methodologies, making results impossible to compare. If companies adopted third-party accreditation systems to assess how they use AI, they would help ensure the accountability that self-reported claims cannot match.

By contrast, there was reasonable progress in this respect as companies adopted ESG principles. For example, institutions such as the Carbon Disclosure Project and Global Reporting Initiative have a network of partners that offer independent verification. These providers, certified under international standards, verify corporate sustainability data against rigorous criteria. That way, they provide the assurance that lets companies show the progress needed to unlock sustainable finance and mitigate legal risks. Third-party audits are far from perfect, but they offer a clear path for improvement.

The fourth principle is robust enforcement. Early ESG initiatives relied on reputational pressure and stakeholder goodwill – things that corporations routinely ignored when profits were at stake. When change came, it was because regulations established legal liability and financial penalties.

These consequences changed how corporations assess risk and continue to shape sustainability practices today. Volkswagen’s 2015 ‘Dieselgate’ scandal, for example, cost the company over US$30 billion in fines, settlements and criminal charges after U.S regulators found that the carmaker was cheating emissions tests. BP faced billions in penalties and liabilities for the 2010 Deepwater Horizon disaster, the biggest oil spill in the history of marine oil drilling operations.

The current enforcement gap in AI creates a predictable dynamic. The expected value of AI washing – like potential investment gains, competitive advantage, and market valuation increases – far exceeds the expected cost in terms of penalties and risk of detection. Until enforcement imposes consequences that exceed benefits, AI washing will persist as a rational business strategy rather than a risk to a business’s reputation.

Fortunately, investors are beginning to step up. The Federal Trade Commission, for example, launched Operation AI Comply in 2024, targeting deceptive AI claims, although this effort has been partially scaled back by the current Trump administration.

New standards for a new era

Until businesses address these four principles, AI washing will continue. Without standards and audits, even well-intentioned companies can’t know if their work meets adequate rigor. Without assessments of material impact, some groups of consumers or shareholders will be hurt. And without liability, even thorough auditors won’t be able to identify whether a business’s claims about AI are truthful.

These principles, applied broadly, also help explain why greenwashing persists. For example, the lack of universal reporting standards continues to leave some gaps, with data-quality issues persisting even as reporting frameworks emerged. More fundamentally, political buy-in for ESG has diminished significantly, particularly in the U.S., where over 150 bills were introduced across multiple states by 2023 to disincentivize firms from adopting ESG. Major financial institutions – including JP Morgan, State Street, BlackRock and PIMCO – have retreated from their earlier climate commitments amid political pressure as well as antitrust concerns.

This trend shows that even well designed accountability measures require durable political support to succeed. After all, corporate sustainability took more than 25 years to develop from an initial framework to mandatory standards, and it still remains a work in progress. AI, by contrast, is advancing exponentially in terms of its reach and societal impact. There may not be 25 years to catch up – but at least there are lessons from the recent past.

Suvrat Dhanorkar, Associate Professor of Operations Management, Georgia Institute of Technology

This article is republished from The Conversation under a Creative Commons license. Read the original article.